Voting patterns in 2016: Exploration using multilevel regression
and poststratification (MRP) on pre-election polls
Rob Trangucci∗‡ Imad Ali† Andrew Gelman† Doug Rivers‡§
01 February 2018
Abstract
We analyzed 2012 and 2016 YouGov pre-election polls in order to understand how
different population groups voted in the 2012 and 2016 elections. We broke the data
down by demographics and state and found:
• The gender gap was an increasing function of age in 2016.
• In 2016 most states exhibited a U-shaped gender gap curve with respect to edu-
cation indicating a larger gender gap at lower and higher levels of education.
• Older white voters with less education more strongly supported Donald Trump
versus younger white voters with more education.
• Women more strongly supported Hillary Clinton than men, with young and more
educated women most strongly supporting Hillary Clinton.
• Older men with less education more strongly supported Donald Trump.
• Black voters overwhelmingly supported Hillary Clinton.
• The gap between college-educated voters and non-college-educated voters was
about 10 percentage points in favor of Hillary Clinton
We display our findings with a series of graphs and maps. The R code associated with
this project is available at https://github.com/rtrangucci/mrp_2016_election/.
∗University of Michigan†Columbia University‡YouGov§Stanford University
1
arX
iv:1
802.
0084
2v3
[st
at.A
P] 1
4 M
ar 2
018
Contents
1 Introduction 3
2 Data and methods 32.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3 Results 63.1 Election results graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1.1 County-level vote swings . . . . . . . . . . . . . . . . . . . . . . . . . 73.1.2 State-level election results and vote swings . . . . . . . . . . . . . . . 10
3.2 Poststratification graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133.2.1 Gender gap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.2.2 Vote by education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183.2.3 Vote by income, age, education, and ethnicity . . . . . . . . . . . . . 193.2.4 Voter turnout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243.2.5 Maps of vote preference . . . . . . . . . . . . . . . . . . . . . . . . . 303.2.6 Maps of voter turnout . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4 Discussion 45
5 Appendix A - Model Code 47
2
1. Introduction
After any election, we typically want to understand how the electorate voted. While nationaland state results give exact measures of aggregate voting, we may be interested in votingbehavior that cuts across state lines, such as how different demographic groups voted. Exitpolls provide one such measure, but without access to the raw data we cannot determineaggregates beyond the margins that are supplied by the exit poll aggregates.
In pursuit of this goal, we can use national pre-election polls in which respondents areasked for whom they plan to vote and post-election polls in which respondents are askedif they participated in the election, both of which record demographic information andstate residency of respondents. Using this data, we then build a statistical model that usesdemographics and state information to predict the probability that an eligible voter votedin the election and which candidate a voter supports. A model that accurately predictsvoting intentions for specific demographic groups (e.g. college-educated Hispanic men livingin Georgia) will require deep interactions as outlined in [1]. In order to precisely learn thesecond- and third-order interactions, we require a large dataset that covers many disparategroups.
Armed with our two models, we can use U.S. Census data to yield the number of peoplein each demographic group. For each group, we then predict the number of voters, and thenumber of votes for each candidate to yield a fine-grained dataset. We can then aggregatethis dataset along any demographic axes we choose in order to investigate voting behavior.
2. Data and methods
2.1. Data
We use YouGov’s daily tracking polls from 10/24/2016 through 11/6/2016 to train the 2016voter preference model. We included 56,946 respondents in the final dataset after filteringout incomplete cases. To train the 2012 voter preference model we used 18,716 respondentspolled on 11/4/2012 from YouGov’s daily tracking poll.
In order to train the 2016 voter turnout model, we use the Current Population Survey(CPS) from 2016, which includes a voting supplement ([2]). The model used 80,766 responsesfrom voters as to whether they voted in the 2016 presidential election. We used the CPSfrom 2012 to train the 2012 voter turnout model, which comprises 81,017 voters. We decidedto use the CPS to train our model because it is viewed as the gold-standard in voter-turnoutpolling [3].
We use a modified version of the 2012 Public Use Microdata Sample Census dataset(PUMS) to get a measure of the total number of eligible voters in the U.S. YouGov providedthe PUMS dataset with ages and education adjusted to match the 2016 population.
2.2. Methods
Our methodology follows that outlined in [4], [1], and [3]. For voter i in group g as definedby the values of a collection of categorical variables, we want to learn the voter’s propensity
3
to vote and for whom they plan to vote, by using a nonrandom sample from the populationof interest. We assume that an individual voter’s response in group g is modeled as follows:
Ti ∼ Bernoulli(αg[i])
where Ti is 1 if the voter plans to vote for Trump, or 0 otherwise. αg[i] is the probabilityof voting for Trump for voter i in group g. In order to make inferences about αg[i] with-out modeling the selection process, we need to stratify our respondents into small enoughgroups so that within a cell selection is random (i.e. that the responses are Bernoulli randomvariables conditional only on g). We do so by generating multidimensional cells defined bydemographic variables like age, ethnicity, and state of residence that categorize our respon-dents. This induces data sparsity even in large polls so we must use Bayesian hierarchicalmodels to partially pool cells along these demographic axes.
Upon fitting our model, we can use the posterior mean of αg, α̂g and Census data to
estimate an aggregate Trump vote proportion by calculating the weighted average∑
g∈DNgα̂g
ND
for whatever demographic category D we like.We measure our electorate using six categorical variables:
• State residency
• Ethnicity
• Gender
• Marital status
• Age
• Education
Each variable v has Lv levels. State residency has fifty levels. Ethnicity has four levels:Black, Hispanic, Other, and White. Gender has two levels. Marital status has three levels:Never married, Married, Not married. Age has four levels, corresponding to the left-closedintervals of age: [18, 30), [30, 45), [45, 65), [65, 99). Education has five levels: No High School,High School, Some College, College, Post Graduate.
After binning our Census data by the six-way interaction of the above attributes, wegenerate table 2.2. Each row of the table represents a specific group of the population, anintersection of six observable attributes. We refer to each row as a cell, and the full table asa six-way poststratification table. Our table has 33,561 cells, reflecting the fact that not allpossible six-way groups exist in the U.S..
We then add columns to this dataset that represent the cell-by-cell probability of votingand the cell-by-cell probability of supporting Trump, which can be combined to yield theexpected number of Trump voters, E [Tg], in each cell g: E [Tg] = N × φg × αg|vote whereφg is the expected probability of voting in cell g, and αg|vote is the expected probability ofvoting for Trump for voters in cell g
In order to generate φg and αg|vote, we build two models: a voter turnout model and avote preference model, respectively. Both models are hierarchical binomial logistic regressionmodels of the form:
4
Table 1: Six-way poststratification tableCell index g State Ethn. Gender . . . Educ. N φg αg|vote E [Tg]
1 AK Black Female . . . College 400 0.40 0.50 802 AK Black Female . . . High School 300 0.30 0.60 54
. . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . .33651 WY White Male . . . Some College 200 0.40 0.40 32
Tg ∼ Binomial(Vg, φg) , g ∈ {1, . . . , G}
logitφg = µ+∑v ∈V
βv[v[g]]
βv ∼ Normal(0, τv)∀v ∈ V
τv =√πv|V |S2
π ∼ Dirichlet(1)
S ∼ Gamma(1, 1)
Each categorical predictor, βv, is represented as a length-Lv vector, where the elements ofthe vector map to the effect associated with the level lv. V denotes the set of all categoricalpredictors included in the model and v[g] is a function that maps the g-th cell to the ap-propriate lv-th level of the categorical predictor. For example, βstate would be a 50-elementvector, and state[ ] is a length-G list of integers with values between 1 and 50 indicating towhich state the g-th cell belongs. Note that the model above can include one-way effects inV , as well as two-way and three-way interactions, like state × age.
We use rstanarm to specify the voter turnout model and the voter preference model,which uses lme4 syntax to facilitate building complex hierarchical generalized linear modelslike above. The full model specifications in lme4 syntax are given in the Appendix. rstanarmimposes more structure on the variance parameters τv than is typical. In our model, τ 2v is theproduct of the square of a global scale parameter S the v-th entry in the simplex parameterπ, and the cardinality of V , |V |. See [5] for more details.
Our voter preference model went through multiple iterations before we arrived at ourfinal model. At first we intended to include past presidential vote. However, PUMS doesnot include past presidential vote, so we used YouGov’s imputed past presidential vote foreach PUMS respondent. This induced too much sparsity in our poststratification frame.
After training each of the models, and generating predictions for voter turnout by celland two-party vote preference for each cell, we adjusted our turnout and vote proportionsin each cell to match the actual state-by-state outcomes as outlined [1].
5
Table 2: Variables in the vote preference model
stan glmer() Variable Description Type Number of Groups
y Vote choice Outcome variable -1 Intercept Global intercept -female Fem.: 0.5, Male: -0.5 Global slope -state pres vote Pre-election poll average Global slope -state State of residence Varying intercept 50age Age Varying intercept 4educ Education attained Varying intercept 51 + state pres vote | eth Ethnicity Varying intercept and slope 4marstat Marital status Varying intercept 3marstat:age Varying intercept 3×4 = 12marstat:state Varying intercept 3×50 = 150marstat:eth Varying intercept 3×4 = 12marstat:gender Varying intercept 3×2 = 6marstat:educ Varying intercept 3×5 = 15state:gender Varying intercept 50×2 = 100age:gender Varying intercept 4×2 = 8educ:gender Varying intercept 5×2 = 10eth:gender Varying intercept 4×2 = 8state:eth Varying intercept 50×4 = 200state:age Varying intercept 50×4 = 200state:educ Varying intercept 50×5 = 250eth:age Varying intercept 4×4 = 16eth:educ Varying intercept 4×5 = 20age:educ Varying intercept 4×5 = 20state:educ:age Varying intercept 50×4×4 = 800educ:age:gender Varying intercept 5×4×2 = 40
3. Results
This section presents plots at the county and state level, followed by charts and maps thatillustrate the poststratification. In addition to vote intention, the charts and maps alsoillustrate voter turnout. The county and state level plots use 2016 and 2012 election resultsand 2010 US census data. The captions of the charts and maps identify which model is usedto produce the data illustrated in the figure. The models are defined as follows:
Model 1 is described in Section 2 above.
Model 2 is similar to Model 1 but includes income as a factor variable and omits maritalstatus. The 2016 vote turnout model for Model 2 was fitted to 2012 CPS.
3.1. Election results graphs
The graphs that follow present actual election results by county and by state. They are notmodel-based, but rather an examination of the Republican vote proportion swing from 2012
6
to 2016 by county versus various demographic variables measured at the county level.
3.1.1. County-level vote swings
Figure 1: County-level Republican Swing by Income
Notes: The county-level Republican swing is computed as Donald Trump’s 2016 two-party vote share minusMitt Romney’s 2012 two-party vote share. Positive values indicate Trump outperforming Romney, whilenegative values indicate Romney outperforming Trump. The area of each circle is proportional to thenumber of voters in each county. Overall, Trump outperformed Romney in counties with lower medianincome. While Trump mostly outperformed Romney in counties with lower voter turnout, Romney mostlyoutperformed Trump in counties with larger voter turnout.
7
Figure 2: County-level Republican Swing by College Education
Notes: The county-level Republican swing is computed as Donald Trump’s 2016 two-party vote share minusMitt Romney’s 2012 two-party vote share. Positive values indicate Trump outperforming Romney, whilenegative values indicate Romney outperforming Trump. The area of each circle is proportional to thenumber of voters in each county. Overall, Trump outperformed Romney in counties with lower collegeeducation. While Trump mostly outperformed Romney in counties with lower voter turnout, Romney mostlyoutperformed Trump in counties with larger voter turnout.
8
Figure 3: County-level Republican Swing by Region as a Function of Income and CollegeEducation
Notes: The county-level Republican swing is computed as Donald Trump’s 2016 two-party vote share minusMitt Romney’s 2012 two-party vote share. Positive values indicate Trump outperforming Romney, whilenegative values indicate Romney outperforming Trump. The area of each circle is proportional to thenumber of voters in each county. Across all regions there is a trend of Trump outperforming Romney in lowincome counties and counties with lower college education. The trend of Trump performing well in countieswith lower college education is less apparent in western counties.
9
3.1.2. State-level election results and vote swings
Figure 4: Republican Share of the Two-Party Vote 2012-2016
0.3 0.4 0.5 0.6 0.7
0.3
0.4
0.5
0.6
0.7
Nationally, Trump got 2% more of the vote than Romney
Romney share of the two−party vote in 2012
Trum
p sh
are
of th
e tw
o−pa
rty
vote
in 2
016
CA
CO
CTDE
FL
HI
IL
IA
ME
MDMA
MIMNNV
NH
NJNM
NY
OH
OR
PA
RI
VT
VA
WA
WI
AL
AK
AZ
AR
GA
ID
INKS
KY
LAMSMOMT
NE
NC
ND OK
SC
SD
TN
TX
UT
WV
WY
Notes: The state-level Republican share of the two-party vote.States are color coded according to the results of the 2012election. States won by Mitt Romney are in red and stateswon by Barack Obama are in blue. The diagonal line indicatesthat the 2012 and 2016 Republican candidates received identicalshares of the two-party vote. In most states Trump received ahigher share of the two-party vote. Nationally, Trump got 2percent more of the two-party vote than Romney.
10
Figure 5: Republican Swing from 2012 to 2016
0.4 0.5 0.6 0.7
Swing from 2012 to 2016: Lots of variation among states
Romney vote in 2012
(Tru
mp
vote
) −
(R
omne
y vo
te)
−4%
−2%
0%2%
4%6%
8%10
%
CA
COCT
DE
FL
HI
IL
IA
ME
MD
MA
MI
MN
NV
NH
NJ
NM
NY
OH
OR
PA
RI
VT
VA
WA
WI
AL
AK
AZ
AR
GA
ID
IN
KS
KY
LA
MS
MO MT
NE
NC
ND
OKSC
SD
TN
TX
WV
WY
Graph omits Utah, where Trumpdid 13% worse than Romney
Notes: The state-level Republican swing. States are color coded according to theresults of the 2012 election. States won by Mitt Romney are in red and stateswon by Barack Obama are in blue. Positive values indicate Trump outperformingRomney and negative values indicate Romney outperforming Trump. There islots of variation among states with Trump outperforming Romney in most states.
11
Figure 6: Trump’s Actual and Forecasted Vote Share
0.4 0.5 0.6 0.7
0.4
0.5
0.6
0.7
Nationally, Trump got 2% more of the vote than predicted
Poll−based forecast of Trump vote
Act
ual T
rum
p vo
te
CA
CO
CTDE
FL
HI
IL
IA
ME
MDMA
MIMN NV
NH
NJNM
NY
OH
OR
PA
RI
VT
VA
WA
WI
AL
AK
AZ
AR
GA
ID
INKS
KY
LAMSMO
MT
NE
NC
ND OK
SC
SD
TN
TX
UT
WV
WY
Notes: A state-level comparison between Donald Trump’s actualtwo-party vote share and his forecasted vote share. States arecolor coded according to the results of the 2012 election. Stateswon by Mitt Romney are in red and states won by BarackObama are in blue. Values on the diagonal indicate that Trump’sactual performance was in line with his forecast. In most statesTrump outperformed his poll-based forecast.
12
Figure 7: Trump’s Actual Minus Forecasted Vote Share
0.4 0.5 0.6 0.7
Trump did much better than predicted in states that Romney won in 2012
Poll−based forecast of Trump vote
(Tru
mp
vote
) −
(P
oll−
base
d fo
reca
st)
−2%
0%2%
4%6%
8%
CA
COCT
DE
FL
HI
IL
IAMEMD
MA
MI
MN
NV
NH
NJ
NM
NY
OH
OR
PA
RI
VT
VA
WA
WIAL
AK
AZ
AR
GA
ID
IN
KS KY
LA
MS
MO
MT
NE
NC
ND
OKSC
SD
TN
TX
UT
WV
WY
Notes: A state-level comparison of Donald Trump’s actual vote share againsthis poll-based forecast. States are color coded according to the results of the2012 election. States won by Mitt Romney are in red and states won by BarackObama are in blue. Positive values indicate states in which Trump outperformedhis forecast and negative values indicate in which Trump’s actual performancefell behind his forecast. Trump did better than predicted in states that Romneywon in 2012.
3.2. Poststratification graphs
The graphs that follow are generated using the multilevel regression and poststratificationmethod outlined in the Methodology section.
13
3.2.1. Gender gap
Figure 8: Gender Gap (Men minus Women) by Education and AgeGender Gap by Education
Gen
der
Gap
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
Gender Gap by Age
18−29 30−44 45−64 65+
0%
5%
10%
15%
20%
Red States: AK TX LA MS AL GA SC TN AR OK KS MO KY WV OH IN IA SD ND WY MT ID UT
Battleground States: ME NH PA NC FL MI WI MN NE CO AZ NV
Blue States: HI VT MA RI CT NJ DE MD NY VA IL NM CA OR WA
Post Strat: pstrat_2016_modeled.RDSNotes: The gender gap is evaluated as men’s probability of voting for Trumpminus women’s probability for of voting for Trump for various education and agelevels. Larger values indicate a greater divergence in vote preference betweenmen and women.(Using Model 1.)
Figure 9: Gender Gap (Men minus Women) by Education and Age - 2012 ElectionGender Gap by Education
Gen
der
Gap
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
Gender Gap by Age
18−29 30−44 45−64 65+
0%
5%
10%
15%
20%
Red States: AK TX LA MS AL GA SC TN AR OK KS MO KY WV OH IN IA SD ND WY MT ID UT
Battleground States: ME NH PA NC FL MI WI MN NE CO AZ NV
Blue States: HI VT MA RI CT NJ DE MD NY VA IL NM CA OR WA
Post Strat: pstrat_2012_modeled.RDSNotes: The gender gap is evaluated as men’s probability of voting for Romneyminus women’s probability for of voting for Romney for various education andage levels.(Using Model 1 with 2012 election results/turnout data.)
14
Figure 10: Gender Gap by Education for each Age Category
Under 30G
ende
r G
ap
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
30−45
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
45−65
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
65+
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
Red States: AK TX LA MS AL GA SC TN AR OK KS MO KY WV OH IN IA SD ND WY MT ID UTBattleground States: ME NH PA NC FL MI WI MN NE CO AZ NVBlue States: HI VT MA RI CT NJ DE MD NY VA IL NM CA OR WA
Post Strat: pstrat_2016_modeled.RDSNotes: The gender gap is evaluated as men’s probability of voting for Trump minus women’s probabil-ity for of voting for Trump for various education levels. Larger values indicate a greater divergence invote preference among women and men. Interactions exist between age and education conditional ongender. Overall, the gender gap increases with age. Among voters under 45 the gender gap is lowestfor those with a college education, and among voters 45 years or older the gender gap is lowest forthose with a high school education.(Using Model 1.)
Figure 11: Gender Gap by Education for each Age Category - 2012 Election
Under 30
Gen
der
Gap
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
30−45
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
45−65
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
65+
< HS HS CollegeSomeCollege
Post−grad
0%
5%
10%
15%
20%
Red States: AK TX LA MS AL GA SC TN AR OK KS MO KY WV OH IN IA SD ND WY MT ID UTBattleground States: ME NH PA NC FL MI WI MN NE CO AZ NVBlue States: HI VT MA RI CT NJ DE MD NY VA IL NM CA OR WA
Post Strat: pstrat_2012_modeled.RDSNotes: The gender gap is evaluated as men’s probability of voting for Romney minus women’s proba-bility for of voting for Romney for various education levels. Larger values indicate a greater divergencein vote preference among women and men. Interactions exist between age and education conditionalon gender.(Using Model 1 with 2012 election results/turnout data.)
15
Figure 12: Gender Gap by Education (Men minus Women)Gender Gap by Education (Men minus Women)
Hawaii
0%
11%
22%California Vermont Massachusetts Maryland New York Illinois Washington Rhode Island
New Jersey
0%
11%
22%Connecticut Oregon Delaware New Mexico Virginia Colorado Maine Nevada
Minnesota
0%
11%
22%New Hampshire Michigan Pennsylvania Wisconsin Florida Arizona North Carolina Georgia
Ohio
0%
11%
22%Texas Iowa South Carolina Alaska Mississippi Missouri Indiana Louisiana
Montana
0%
11%
22%Kansas Utah Nebraska Tennessee Arkansas
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Alabama
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Kentucky
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
South Dakota
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Idaho
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
0%
11%
22%Oklahoma
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
North Dakota
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
West Virginia
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Wyoming
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Post Strat: pstrat_2016_modeled.RDS
Notes: The state-level gender gap is evaluated as men’s probability of voting for Trump minus women’sprobability for of voting for Trump for various education levels. Larger values indicate a greater divergencein vote preference among women and men. In most states, voters with a high school education level tend tohave the lowest gender gap and voters with a post graduate education level tend to have the highest gendergap.(Using Model 1.)
16
Figure 13: Gender Gap by Age (Men minus Women)Gender Gap by Age (Men minus Women)
Hawaii
0%
10%
20%California Vermont Massachusetts Maryland New York Illinois Washington Rhode Island
New Jersey
0%
10%
20%Connecticut Oregon Delaware New Mexico Virginia Colorado Maine Nevada
Minnesota
0%
10%
20%New Hampshire Michigan Pennsylvania Wisconsin Florida Arizona North Carolina Georgia
Ohio
0%
10%
20%Texas Iowa South Carolina Alaska Mississippi Missouri Indiana Louisiana
Montana
0%
10%
20%Kansas Utah Nebraska Tennessee Arkansas
Und
er 3
0
30−
45
45−
65
65+
Alabama
Und
er 3
0
30−
45
45−
65
65+
Kentucky
Und
er 3
0
30−
45
45−
65
65+
South Dakota
Und
er 3
0
30−
45
45−
65
65+
Idaho
Und
er 3
0
30−
45
45−
65
65+
0%
10%
20%Oklahoma
Und
er 3
0
30−
45
45−
65
65+
North Dakota
Und
er 3
0
30−
45
45−
65
65+
West Virginia
Und
er 3
0
30−
45
45−
65
65+
Wyoming
Und
er 3
0
30−
45
45−
65
65+
Post Strat: pstrat_2016_modeled.RDS
Notes: The state-level gender gap is evaluated as men’s probability of voting for Trump minus women’sprobability for of voting for Trump for various education levels. Larger values indicate a greater divergencein vote preference among women and men. The gender gap increases with age in most states, with largervariation in states that supported Clinton.(Using Model 1.)
17
3.2.2. Vote by education
Figure 14: Trump’s Share of the Two-Party Vote by Education for each Age Category
Under 30
Trum
p S
hare
of T
wo−
Par
ty V
ote
< HS HS CollegeSomeCollege
Post−grad
10%
20%
30%
40%
50%
60%
70%
80%
30−45
< HS HS CollegeSomeCollege
Post−grad
10%
20%
30%
40%
50%
60%
70%
80%
45−65
< HS HS CollegeSomeCollege
Post−grad
10%
20%
30%
40%
50%
60%
70%
80%
65+
< HS HS CollegeSomeCollege
Post−grad
10%
20%
30%
40%
50%
60%
70%
80%
Red States: AK TX LA MS AL GA SC TN AR OK KS MO KY WV OH IN IA SD ND WY MT ID UTBattleground States: ME NH PA NC FL MI WI MN NE CO AZ NVBlue States: HI VT MA RI CT NJ DE MD NY VA IL NM CA OR WA
Post Strat: pstrat_2016_modeled.RDSNotes: Republican share of the two-party vote against various education levels. Overall, the Republi-can share increases with age. The strongest support came from voters with a high school education ineach age category, with the exception of 30-45 year olds.(Using Model 1.)
Figure 15: Romney’s Share of the Two-Party Vote by Education for each Age Category -2012 Election
Under 30
Rom
eny
Sha
re o
f Tw
o−P
arty
Vot
e
< HS HS CollegeSomeCollege
Post−grad
10%
20%
30%
40%
50%
60%
70%
80%
30−45
< HS HS CollegeSomeCollege
Post−grad
10%
20%
30%
40%
50%
60%
70%
80%
45−65
< HS HS CollegeSomeCollege
Post−grad
10%
20%
30%
40%
50%
60%
70%
80%
65+
< HS HS CollegeSomeCollege
Post−grad
10%
20%
30%
40%
50%
60%
70%
80%
Red States: AK TX LA MS AL GA SC TN AR OK KS MO KY WV OH IN IA SD ND WY MT ID UTBattleground States: ME NH PA NC FL MI WI MN NE CO AZ NVBlue States: HI VT MA RI CT NJ DE MD NY VA IL NM CA OR WA
Post Strat: pstrat_2012_modeled.RDSNotes: Republican share of the two-party vote against various education levels. Overall, the Republi-can share increases with age.(Using Model 1 with 2012 election results/turnout data.)
18
3.2.3. Vote by income, age, education, and ethnicity
Figure 16: Trump’s Share of the Two-Party Vote by Income and EducationTrump's Share of Vote by Income
0%
25%
50%
75%U
nder
$30
k
$30−
50k
$50−
100k
Ove
r $1
00k
Trum
p's
Sha
re o
f Vot
e
Post Strat: pstrat_income_UGov_wave_20161130.RDS
Trump's Share of Vote by Education
0%
25%
50%
75%
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Trum
p's
Sha
re o
f Vot
e
Post Strat: pstrat_2016_modeled.RDS
WhitesBlacksHispanicsOthersOverall
Notes: Republican share of the two-party vote for Whites (orange), Blacks (black), Hispan-ics (red), other ethnicities (green), and overall (blue). Trump’s share of the vote is highestamong white voters with a high school education level.(Using Model 2 (left) and Model 1 (right).)
19
Figure 17: Trump’s Share of the Two-Party Vote by Education, Ethnicity, and StateTrump's Share of Vote by Education
Hawaii
0%
50%
100%California Vermont Massachusetts Maryland New York Illinois Washington Rhode Island
New Jersey
0%
50%
100%Connecticut Oregon Delaware New Mexico Virginia Colorado Maine Nevada
Minnesota
0%
50%
100%New Hampshire Michigan Pennsylvania Wisconsin Florida Arizona North Carolina Georgia
Ohio
0%
50%
100%Texas Iowa South Carolina Alaska Mississippi Missouri Indiana Louisiana
Montana
0%
50%
100%Kansas Utah Nebraska Tennessee Arkansas
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Alabama
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Kentucky
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
South Dakota
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Idaho
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
0%
50%
100%Oklahoma
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
North Dakota
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
West Virginia
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Wyoming
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
WhiteBlack
HispanicOther
Overall
Post Strat: pstrat_2016_modeled.RDS
Notes: State-level Republican share of the two-party vote for Whites (orange), Blacks (black), Hispanics(red), other ethnicities (green), and overall (blue). In most states white voters with high school educationhave the greatest support for Trump and those with post graduate education have the lowest support forTrump.(Using Model 1.)
20
Figure 18: Romney’s Share of the Two-Party Vote by Education, Ethnicity, and State - 2012Election Romeny's Share of Vote by Education
Hawaii
0%
50%
100%Vermont New York Rhode Island Maryland California Massachusetts Delaware New Jersey
Connecticut
0%
50%
100%Illinois Maine Washington Oregon New Mexico Michigan Minnesota Wisconsin
Nevada
0%
50%
100%Iowa New Hampshire Colorado Pennsylvania Virginia Ohio Florida North Carolina
Georgia
0%
50%
100%Arizona Missouri Indiana South Carolina Mississippi Montana Alaska Texas
Louisiana
0%
50%
100%South Dakota North Dakota Tennessee Kansas Nebraska
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Alabama
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Kentucky
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Arkansas
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
West Virginia
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
0%
50%
100%Idaho
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Oklahoma
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Wyoming
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Utah
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
WhiteBlack
HispanicOther
Overall
Post Strat: pstrat_2012_modeled.RDS
Notes: State-level Republican share of the two-party vote for Whites (orange), Blacks (black), Hispanics(red), other ethnicities (green), and overall (blue).(Using Model 1 with 2012 election results/turnout data.)
21
Figure 19: Trump’s Share of the Two-Party Vote by Age, Ethnicity, and StateTrump's Share of Vote by Age
Hawaii
0%
50%
100%California Vermont Massachusetts Maryland New York Illinois Washington Rhode Island
New Jersey
0%
50%
100%Connecticut Oregon Delaware New Mexico Virginia Colorado Maine Nevada
Minnesota
0%
50%
100%New Hampshire Michigan Pennsylvania Wisconsin Florida Arizona North Carolina Georgia
Ohio
0%
50%
100%Texas Iowa South Carolina Alaska Mississippi Missouri Indiana Louisiana
Montana
0%
50%
100%Kansas Utah Nebraska Tennessee Arkansas
Und
er 3
0
30−
45
45−
65
65+
Alabama
Und
er 3
0
30−
45
45−
65
65+
Kentucky
Und
er 3
0
30−
45
45−
65
65+
South Dakota
Und
er 3
0
30−
45
45−
65
65+
Idaho
Und
er 3
0
30−
45
45−
65
65+
0%
50%
100%Oklahoma
Und
er 3
0
30−
45
45−
65
65+
North Dakota
Und
er 3
0
30−
45
45−
65
65+
West Virginia
Und
er 3
0
30−
45
45−
65
65+
Wyoming
Und
er 3
0
30−
45
45−
65
65+
WhiteBlack
HispanicOther
Overall
Post Strat: pstrat_2016_modeled.RDS
Notes: State-level Republican share of the two-party vote for Whites (orange), Blacks (black), Hispanics(red), other ethnicities (green), and overall (blue). Support for Trump increases with age. Support amongWhites is consistently the strongest followed by support among other races and Hispanics.(Using Model 1.)
22
Figure 20: Romney’s Share of the Two-Party Vote by Age, Ethnicity, and State - 2012Election Romeny's Share of Vote by Age
Hawaii
0%
50%
100%Vermont New York Rhode Island Maryland California Massachusetts Delaware New Jersey
Connecticut
0%
50%
100%Illinois Maine Washington Oregon New Mexico Michigan Minnesota Wisconsin
Nevada
0%
50%
100%Iowa New Hampshire Colorado Pennsylvania Virginia Ohio Florida North Carolina
Georgia
0%
50%
100%Arizona Missouri Indiana South Carolina Mississippi Montana Alaska Texas
Louisiana
0%
50%
100%South Dakota North Dakota Tennessee Kansas Nebraska
Und
er 3
0
30−
45
45−
65
65+
Alabama
Und
er 3
0
30−
45
45−
65
65+
Kentucky
Und
er 3
0
30−
45
45−
65
65+
Arkansas
Und
er 3
0
30−
45
45−
65
65+
West Virginia
Und
er 3
0
30−
45
45−
65
65+
0%
50%
100%Idaho
Und
er 3
0
30−
45
45−
65
65+
Oklahoma
Und
er 3
0
30−
45
45−
65
65+
Wyoming
Und
er 3
0
30−
45
45−
65
65+
Utah
Und
er 3
0
30−
45
45−
65
65+
WhiteBlack
HispanicOther
Overall
Post Strat: pstrat_2012_modeled.RDS
Notes: State-level Republican share of the two-party vote for Whites (orange), Blacks (black), Hispanics(red), other ethnicities (green), and overall (blue). Support for Trump increases with age.(Using Model 1 with 2012 election results/turnout data.)
23
3.2.4. Voter turnout
Figure 21: Voter Turnout by Education, Ethnicity and StateVoter Turnout by Education
Hawaii
10%
50%
90%California Vermont Massachusetts Maryland New York Illinois Washington Rhode Island
New Jersey
10%
50%
90%Connecticut Oregon Delaware New Mexico Virginia Colorado Maine Nevada
Minnesota
10%
50%
90%New Hampshire Michigan Pennsylvania Wisconsin Florida Arizona North Carolina Georgia
Ohio
10%
50%
90%Texas Iowa South Carolina Alaska Mississippi Missouri Indiana Louisiana
Montana
10%
50%
90%Kansas Utah Nebraska Tennessee Arkansas
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Alabama
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Kentucky
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
South Dakota
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Idaho
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
10%
50%
90%Oklahoma
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
North Dakota
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
West Virginia
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Wyoming
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
WhiteBlack
HispanicOther
Overall
Post Strat: pstrat_2016_modeled.RDS
Notes: Voter turnout for Whites (orange), Blacks (black), Hispanics (red), other ethnicities (green), andoverall (blue). Voter turnout increases with education. There is not much variation across states. Withinstates Hispanics typically experienced low voter turnout compared to Whites and Blacks.(Using Model 1.)
24
Figure 22: Voter Turnout by Education, Ethnicity and State - 2012 ElectionVoter Turnout by Education
Hawaii
10%
50%
90%Vermont New York Rhode Island Maryland California Massachusetts Delaware New Jersey
Connecticut
10%
50%
90%Illinois Maine Washington Oregon New Mexico Michigan Minnesota Wisconsin
Nevada
10%
50%
90%Iowa New Hampshire Colorado Pennsylvania Virginia Ohio Florida North Carolina
Georgia
10%
50%
90%Arizona Missouri Indiana South Carolina Mississippi Montana Alaska Texas
Louisiana
10%
50%
90%South Dakota North Dakota Tennessee Kansas Nebraska
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Alabama
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Kentucky
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Arkansas
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
West Virginia
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
10%
50%
90%Idaho
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Oklahoma
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Wyoming
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Utah
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
WhiteBlack
HispanicOther
Overall
Post Strat: pstrat_2012_modeled.RDS
Notes: Voter turnout for Whites (orange), Blacks (black), Hispanics (red), other ethnicities (green), andoverall (blue).(Using Model 1 with 2012 election results/turnout data.)
25
Figure 23: Voter Turnout by Age, Ethnicity and StateVoter Turnout by Age
Hawaii
15%
50%
85% California Vermont Massachusetts Maryland New York Illinois Washington Rhode Island
New Jersey
15%
50%
85% Connecticut Oregon Delaware New Mexico Virginia Colorado Maine Nevada
Minnesota
15%
50%
85% New Hampshire Michigan Pennsylvania Wisconsin Florida Arizona North Carolina Georgia
Ohio
15%
50%
85% Texas Iowa South Carolina Alaska Mississippi Missouri Indiana Louisiana
Montana
15%
50%
85% Kansas Utah Nebraska Tennessee Arkansas
Und
er 3
0
30−
45
45−
65
65+
Alabama
Und
er 3
0
30−
45
45−
65
65+
Kentucky
Und
er 3
0
30−
45
45−
65
65+
South Dakota
Und
er 3
0
30−
45
45−
65
65+
Idaho
Und
er 3
0
30−
45
45−
65
65+
15%
50%
85% Oklahoma
Und
er 3
0
30−
45
45−
65
65+
North Dakota
Und
er 3
0
30−
45
45−
65
65+
West Virginia
Und
er 3
0
30−
45
45−
65
65+
Wyoming
Und
er 3
0
30−
45
45−
65
65+
WhiteBlack
HispanicOther
Overall
Post Strat: pstrat_2016_modeled.RDS
Notes: Voter turnout for Whites (orange), Blacks (black), Hispanics (red), other ethnicities (green), andoverall (blue). Voter turnout increases with age. There is low voter turnout among Hispanics across agelevels compared to Whites and Blacks.(Using Model 1.)
26
Figure 24: Voter Turnout by Age, Ethnicity and State - 2012 ElectionVoter Turnout by Age
Hawaii
15%
50%
85% Vermont New York Rhode Island Maryland California Massachusetts Delaware New Jersey
Connecticut
15%
50%
85% Illinois Maine Washington Oregon New Mexico Michigan Minnesota Wisconsin
Nevada
15%
50%
85% Iowa New Hampshire Colorado Pennsylvania Virginia Ohio Florida North Carolina
Georgia
15%
50%
85% Arizona Missouri Indiana South Carolina Mississippi Montana Alaska Texas
Louisiana
15%
50%
85% South Dakota North Dakota Tennessee Kansas Nebraska
Und
er 3
0
30−
45
45−
65
65+
Alabama
Und
er 3
0
30−
45
45−
65
65+
Kentucky
Und
er 3
0
30−
45
45−
65
65+
Arkansas
Und
er 3
0
30−
45
45−
65
65+
West Virginia
Und
er 3
0
30−
45
45−
65
65+
15%
50%
85% Idaho
Und
er 3
0
30−
45
45−
65
65+
Oklahoma
Und
er 3
0
30−
45
45−
65
65+
Wyoming
Und
er 3
0
30−
45
45−
65
65+
Utah
Und
er 3
0
30−
45
45−
65
65+
WhiteBlack
HispanicOther
Overall
Post Strat: pstrat_2012_modeled.RDS
Notes: Voter turnout for Whites (orange), Blacks (black), Hispanics (red), other ethnicities (green), andoverall (blue).(Using Model 1 with 2012 election results/turnout data.)
27
Figure 25: Voter Turnout by Education, Gender and StateVoter Turnout by Education
Hawaii
10%
50%
90%California Vermont Massachusetts Maryland New York Illinois Washington Rhode Island
New Jersey
10%
50%
90%Connecticut Oregon Delaware New Mexico Virginia Colorado Maine Nevada
Minnesota
10%
50%
90%New Hampshire Michigan Pennsylvania Wisconsin Florida Arizona North Carolina Georgia
Ohio
10%
50%
90%Texas Iowa South Carolina Alaska Mississippi Missouri Indiana Louisiana
Montana
10%
50%
90%Kansas Utah Nebraska Tennessee Arkansas
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Alabama
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Kentucky
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
South Dakota
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Idaho
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
10%
50%
90%Oklahoma
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
North Dakota
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
West Virginia
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Wyoming
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Women Men Overall
Post Strat: pstrat_2016_modeled.RDS
Notes: Voter turnout for women (red), men (blue), and overall (grey). Voter turnout increases witheducation, with women experiencing a larger voter turnout compared to men.(Using Model 1.)
28
Figure 26: Voter Turnout by Education, Gender and State - 2012 ElectionVoter Turnout by Education
Hawaii
10%
50%
90%Vermont New York Rhode Island Maryland California Massachusetts Delaware New Jersey
Connecticut
10%
50%
90%Illinois Maine Washington Oregon New Mexico Michigan Minnesota Wisconsin
Nevada
10%
50%
90%Iowa New Hampshire Colorado Pennsylvania Virginia Ohio Florida North Carolina
Georgia
10%
50%
90%Arizona Missouri Indiana South Carolina Mississippi Montana Alaska Texas
Louisiana
10%
50%
90%South Dakota North Dakota Tennessee Kansas Nebraska
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Alabama
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Kentucky
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Arkansas
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
West Virginia
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
10%
50%
90%Idaho
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Oklahoma
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Wyoming
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Utah
No
Hig
h S
choo
l
Hig
h S
choo
l
Som
e C
olle
ge
Col
lege
Pos
t Gra
duat
e
Women Men Overall
Post Strat: pstrat_2012_modeled.RDS
Notes: Voter turnout for women (red), men (blue), and overall (grey).(Using Model 1 with 2012 election results/turnout data.)
29
3.2.5. Maps of vote preference
Figure 27: Gender Gap (Men minus Women)
Notes: State-level gender gap evaluated as men’s probability of voting for Trump minus women’s probabilityfor of voting for Trump. Dark green/orange indicates a larger divergence in vote preference between menand women. The greatest divergence exists among older voters with post graduate education. The weakestsupport exists among young voters with a college education.(Using Model 1.)
30
Figure 28: Trump’s Share of the Two-Party Vote by Age and Education
Notes: State-level vote intention by education and age. Dark red indicates stronger support for DonaldTrump and dark blue indicates stronger support for Hillary Clinton. Overall, older voters with lowereducation have stronger support for Trump and younger voters with higher levels of education have strongersupport for Clinton. In each age bracket Trump has stronger support among voters with high school andsome college education compared to voters with no high school education.(Using Model 1.)
31
Figure 29: Trump’s Share of the Two-Party Vote by Age and Education for Women
Notes: State-level vote intention by education and age for women. Dark red indicates stronger support forDonald Trump and dark blue indicates stronger support for Hillary Clinton. Overall, older women havestronger support for Trump. Women with a post graduate education have stronger support for Clinton,and women with a high school education and some college education have stronger support for Trump.(Using Model 1.)
32
Figure 30: Trump’s Share of the Two-Party Vote by Age and Education for Men
Notes: State-level vote intention by education and age for men. Dark red indicates stronger support forDonald Trump and dark blue indicates stronger support for Hillary Clinton. Older men have strongersupport for Trump whereas younger men have stronger support for Clinton. Overall, men with a postgraduate education have stronger support for Clinton, while men with a high school education have strongersupport for Trump.(Using Model 1.)
33
Figure 31: Trump’s Share of the Two-Party Vote by Age and Education for Whites
Notes: State-level vote intention by education and age for Whites. Dark red indicates stronger supportfor Donald Trump and dark blue indicates stronger support for Hillary Clinton. Older voters with lesseducation had stronger support for Trump, whereas younger voters with more education had strongersupport for Clinton. In terms of education, the strongest support for Clinton comes from voters with a postgraduate education and the strongest support for Trump comes from voters with a high school education.(Using Model 1.)
34
Figure 32: Trump’s Share of the Two-Party Vote by Age and Education for Blacks
Notes: State-level vote intention by education and age for Blacks. Dark red indicates stronger support forDonald Trump among women and dark blue indicates stronger support for Hillary Clinton. Missing cellsare denoted by diagonal lines. Overall, Blacks supported Clinton.(Using Model 1.)
35
Figure 33: Trump’s Share of the Two-Party Vote by Age and Education for Hispanics
Notes: State-level vote intention by education and age for Hispanics. Dark red indicates stronger supportfor Donald Trump and dark blue indicates stronger support for Hillary Clinton. Missing cells are denotedby diagonal lines. A majority of young Hispanics have stronger support for Clinton. Support for Trumpincreases with age at all education levels. There is not much variation across education levels.(Using Model 1.)
36
Figure 34: Trump’s Share of the Two-Party Vote by Age and Education for Other Ethnicities
Notes: State-level vote intention by education and age for ethnicities (not including White, Black, orHispanic). Dark red indicates stronger support for Donald Trump and dark blue indicates stronger supportfor Hillary Clinton. Support for Trump increases with age at all education levels. Support for Trumpconsistently decreases with education (with the exception of the 65+ age bracket).(Using Model 1.)
37
Figure 35: Trump’s Share of the Two-Party Vote by Education and White vs. Non-white
Notes: State-level vote intention for white and non-white voters by education.No college education includes the categories “No High School”, “High School”,and “Some College”. College education includes the categories “College” and“Post Graduate”. Dark red indicates stronger support for Donald Trump anddark blue indicates stronger support for Hillary Clinton. White voters havestronger support for Trump compared to non-white voters, with white voterswith no college education having the strongest support. There is little variationin vote preference across these categories for North Dakota, Wyoming, andIdaho, which consistently support Trump. There is also little variation in votepreference across education levels among non-white voters.(Using Model 1.)
38
Figure 36: Romney’s Share of the Two-Party Vote by Education and White vs. Non-white- 2012 Election
Notes: State-level vote intention for white and non-white voters by educa-tion. No college education includes the categories “No High School”, “HighSchool”, and “Some College”. College education includes the categories“College” and “Post Graduate”. Dark red indicates stronger supportfor Mitt Romney and dark blue indicates stronger support for BarackObama. White voters with no college education had the strongest supportfor Romney. Regardless of college education, non-White voters had thestrongest support for Obama.(Using Model 1 with 2012 election results/turnout data.)
39
Figure 37: Trump’s Share of the Two-Party Vote by Education and White vs. Non-whiteWomen
Notes: State-level vote intention for white and non-white women by education. Dark red indicates strongersupport for Donald Trump among women and dark blue indicates stronger support for Hillary Clintonamong women. Support for Trump among white women increases from no high school to high schooleducation levels and declines from high school to post graduate education levels. White women with highschool education have the strongest support for Trump. Overall, non-white women have stronger supportfor Clinton, with the exception of some Midwestern states (e.g. North Dakota and Wyoming).(Using Model 1.)
Figure 38: Romney’s Share of the Two-Party Vote by Education and White vs. Non-whiteWomen - 2012 Election
Notes: State-level vote intention for white and non-white women by education. Dark red indicates strongersupport for Mitt Romney among women and dark blue indicates stronger support for Barack Obamaamong women. Support for Romney among White women decreased with education. Regardless of collegeeducation, Obama had strong support among non-White women.(Using Model 1 with 2012 election results/turnout data.)
40
Figure 39: Trump’s Share of the Two-Party Vote by Education for Women
Notes: State-level vote intention for women by education. Dark red indicates stronger support for DonaldTrump among women and dark blue indicates stronger support for Hillary Clinton among women. In moststates, women with high school education have stronger support for Trump and women with post graduateeducation have stronger support for Clinton.(Using Model 1.)
Figure 40: Romney’s Share of the Two-Party Vote by Education for Women - 2012 Election
Notes: State-level vote intention for women by education. Dark red indicates stronger support for MittRomney among women and dark blue indicates stronger support for Barack Obama among women. In moststates, women with high school education had stronger support for Romney and women with post graduateeducation had stronger support for Obama.(Using Model 1 with 2012 election results/turnout data.)
41
3.2.6. Maps of voter turnout
Figure 41: Voter Turnout by Age and Education
Notes: State-level voter turnout by education and age. Yellow indicates low voter turnout and dark blueindicates high voter turnout. Younger individuals with less education were less likely to vote this election,whereas older individuals with more education were more likely to vote.(Using Model 1.)
42
Figure 42: Voter Turnout by Age and Education for Women
Notes: State-level voter turnout by education and age for women. Yellow indicates low voter turnout anddark blue indicates high voter turnout.(Using Model 1.)
43
Figure 43: Voter Turnout by Age and Education for Men
Notes: State-level voter turnout by education and age for women. Yellow indicates low voter turnout anddark blue indicates high voter turnout.(Using Model 1.)
44
Figure 44: Voter Turnout Gender Gap (men minus women)
Notes: State-level voter turnout gender gap evaluated as voter turnout probability for men minus voterturnout probability for women. Dark green/orange indicates a large turnout gender gap.(Using Model 1.)
4. Discussion
We keep the discussion short as we feel that our main contribution here is to present thesegraphs and maps which others can interpret how they see best, and to share our code sothat others can fit these and similar models on their own.
Some of our findings comport with the broader media narrative developed in the af-termath of the election. We found that white voters with lower educational attainmentsupported Trump nearly uniformly. We did not find that income was a strong predictor ofsupport for Trump, perhaps a continuation of a trend apparent in 2000 through 2012 electiondata. We found the gender gap to be about 10%, which was a bit lower than predicted byexit polls. The marital status gap we estimated was about 2× the figure estimated by exitpolls.
Most surprising to us was the strong age pattern in the gender gap. Older women weremuch more likely to support Clinton than older men, while younger women were mildly morelikely to support Clinton compared to men the same age. We are not sure what accountsfor this difference. One area of future research is using age as a continuous predictor ratherthan binning ages and using the bins as categorical predictors.
45
Our models predict that men in several state by education categories were more likelyto support Clinton than women. We do not believe this to be true but rather believe itto be a problem with poststratification table sparsity. In order to reduce the number ofpoststratification cells, in future analyses we could poststratify by region rather than state.This would likely not have impacted our descriptive precision in this analysis due to theapparently strong regional patterns in voting behavior in this election.
46
5. Appendix A - Model Code
We specified our voter turnout model as below:
cbind(vote, did_not_vote) ~ 1 + female + state_pres_vote +
(1 | state) + (1 | age) +
(1 | educ) + (1 + state_pres_vote | eth) +
(1 | marstat) + (1 | marstat:age) +
(1 | marstat:state) + (1 | marstat:eth) +
(1 | marstat:gender) + (1 | marstat:educ) +
(1 | state:gender) + (1 | age:gender) +
(1 | educ:gender) + (1 | eth:gender) +
(1 | state:eth) + (1 | state:age) +
(1 | state:educ) + (1 | eth:age) +
(1 | eth:educ) + (1 | age:educ) +
(1 | state:educ:age) + (1 | educ:age:gender)
We specified our voter preference model as below:
cbind(clinton, trump) ~ 1 + female + state_pres_vote +
(1 | state) + (1 | age) +
(1 | educ) + (1 + state_pres_vote | eth) +
(1 | marstat) + (1 | marstat:age) +
(1 | marstat:state) + (1 | marstat:eth) +
(1 | marstat:gender) + (1 | marstat:educ) +
(1 | state:gender) + (1 | age:gender) +
(1 | educ:gender) + (1 | eth:gender) +
(1 | state:eth) + (1 | state:age) +
(1 | state:educ) + (1 | eth:age) +
(1 | eth:educ) + (1 | age:educ) +
(1 | state:educ:age) + (1 | educ:age:gender)
47
References
[1] Yair Ghitza and Andrew Gelman. Deep interactions with MRP: Election turnout andvoting patterns among small electoral subgroups. American Journal of Political Science,57(3):762–776, 2013.
[2] Sarah Flood, Miriam King, Steven Ruggles, and J. Robert Warren. Integrated publicuse microdata series, current population survey: Version 5.0. [dataset], 2017.
[3] Rayleigh Lei, Andrew Gelman, and Yair Ghitza. The 2008 election: A preregisteredreplication analysis. Statistics and Public Policy, 4(1):1–8, 2017.
[4] Andrew Gelman and Thomas C Little. Poststratification into many categories usinghierarchical logistic regression. Survey Methodology, 23(2):127–135, 1997.
[5] Stan Development Team. RStanArm: Bayesian applied regression modeling via Stan. Rpackage version 2.13.1., 2016.
48